On Thu, 10 May 2007, Ulrich Schwickerath wrote:
Hi,
> Hi,
>
> > There is one problem yet, if i have two CE1 CE2 sharing one torque server
> > and if i want to disable a queue only on CE1 , while CE2 would continue
> > to using the queue, the only current way ,i see, to do it is to:
> > modify the /opt/lcg/var/gip/ldif/static-file-CE.ldif file
> > and also temporarily remove lcg-info-dynamic-ce from
> > /opt/lcg/var/gip/plugin.
> >
> > Another possible ways:
> > 1)is to moify the /opt/lcg/libexec/lcg-info-dynamic-pbs
> > to read the GlueCEStateStatus from some configuration file which the admin
> > would change upon need, which would ovveride or avoid quering the enabled
> > & started values from pbs.
> >
> > 2)Or modify /opt/lcg/var/gip/ldif/static-file-CE.ldif and if
> > /opt/lcg/libexec/lcg-info-dynamic-pbs has GlueCEStateStatus != Production,
> > avoid quering pbs , and print the GlueCEStateStatus from
> > static-file-CE.ldif instead.
> >
> > Please correct me if i'm wrong. If i'am correct could the developers
> > be asked to modify /opt/lcg/libexec/lcg-info-dynamic-(pbs|lsf|condor)
> > according to 1 or 2?
>
> I fully agree.
>
> Actually something like that is already implemented in the information
> providers for LSF, version >2.0. They come with a configuration
> file /opt/glite/etc/lcg-info-dynamic-lsf.conf in which one can define a
> parameter LSF_CE_DEFAULT_STATUS. If this parameter is either undefined or set
> to DEFAULT, the information providers will put the CE status to what
> corresponds to the queue status, i.e. draining if a queue schedules jobs but
> does not accept new jobs, closed if the queue is closed and scheduling is
> disabled, queueing if scheduling is disabled but queuing is on. This default
> behaviour can be overriden by setting LSF_CE_DEFAULT_STATUS to one of
> Production
> Queueing
> Draining
> Closed
> At CERN we have a mechanism to put machines into standby or maintenance. If
> the CE node status changes to standby, a small script changes the setting of
> LSF_CE_DEFAULT_STATUS to draining, or closed in case of maintenance.
Thanks for the info.
This schema has a drawback , since it is not implemented on VO
granualrity but on a whole CE, it would be metter to have also
LSF_CE_<VO>_DEFAULT_STATUS, for each VO, in case there is a need
to disable only one queue.
> Maybe this scheme could be implemented for pbs, too.
I don't mind modifying the lcg-info-dynamic-pbs to support similar schema
but only in case it will be accepted by developers, otherwise i don't want
to bother.
Alex
|